Unlocking FUSE: Optimizing Cache Consistency for Faster Cloud‑Native Storage

The 2022 China Cloud Computing Infrastructure Developer Conference (CID) will be held on November 6 in Hangzhou, organized by ByteDance, Alibaba Cloud, Huawei Cloud, Intel, and Tencent Cloud. The event will feature over 30 talks covering networking, storage, virtualization, cloud orchestrators (Kubernetes, etc.), cloud operating systems (OpenStack, etc.), cloud security, cloud firmware, DPU/IPU/smart NICs, and other cloud‑native technologies such as micro‑services, Serverless, and service mesh.

At this conference, ByteDance's System Technology & Engineering (STE) team member Zhang Jiachen will present a session titled "FUSE Cache Consistency Model and Its Optimization" . The FUSE user‑space file system protocol is widely used for secure container host directory passthrough, distributed file system clients, and similar scenarios, where the tension between performance and consistency has long been a focus.

The talk will deeply analyze the root causes of this performance‑consistency conflict, discuss the trade‑offs, and conclude with case studies that demonstrate how to make informed choices and apply optimization recommendations for FUSE consistency.

Topic Details

Topic Title: FUSE Cache Consistency Model and Its Optimization

Speaker Bio: Zhang Jiachen, member of ByteDance's STE team, focuses on storage virtualization, file systems, persistent memory systems, and their design and optimization in virtualized and cloud‑native environments. He has previously spoken at CID 2020, KVM Forum 2021, and other conferences.

Abstract: Performance and consistency are fundamental contradictions in distributed system design: maintaining consistency requires more inter‑node communication, which can degrade performance. The FUSE protocol’s performance‑consistency tension stems mainly from synchronization issues of various client caches. This presentation will first introduce the native FUSE cache mode and consistency model, analyze potential issues, then describe optimizations—including enhancing the high‑performance write‑back cache’s close‑to‑open (CTO) consistency model—and discuss the trade‑offs between multiple consistency models and performance. Finally, a case study will illustrate the necessity of selecting and optimizing the appropriate FUSE consistency model.